I'm sure that kind of stuff requires solving a few LCP's per sample. ;-) Didn't they say solving for the modes in a mesh was about two hours of pre-computation? I think you'd want to do that once per physics frame. Per object ;-) Cheers, / h+